Raiden solves VRP at Rust speed using adaptive large neighborhood search. Pay for the compute seconds your problem uses. Not for routes, not for stops, not for vehicles.
Per-stop and per-route pricing means scaling your delivery operation directly inflates your API bill. Raiden charges for the CPU seconds your solve actually consumed. Nothing more.
1,000 stops/day costs 10× more than 100. Every order you win raises your infrastructure bill.
Solve 100 stops or 10,000: your unit cost is the same. Scale routes, not bills.
Expanding your fleet forces a plan upgrade mid-contract.
No fleet-size limits. Add vans, add shifts. Raiden doesn't count vehicles.
No visibility into solution quality, iteration counts, or why a route looks the way it does.
CVRPLIB and Solomon VRPTW scores are public. Judge quality before you integrate.
Real limits are behind "contact sales." Developers can't prototype without a meeting.
Start prototyping in minutes. No demo required to see real limits.
Last-mile specifics (time windows, service time, committed stops) are afterthoughts.
Time windows, capacity, service duration, committed-stop locking: all first-class.
A clean REST API. No SDK required. POST your stops, GET optimized routes. Every response includes cost, timing, and solve diagnostics.
# Optimize 100 stops across 3 vehicles with time windows curl https://api.raiden.dev/v1/optimize \ -H "Authorization: Bearer $RAIDEN_KEY" \ -H "Content-Type: application/json" \ -d "depot": "lat": 51.5074, "lng": -0.1278 , "vehicles": [ "id": "van_1", "capacity": 800, "shift": ["08:00", "18:00"] ], "stops": [ "id": "order_291", "lat": 51.52, "lng": -0.09, "demand": 12, "window": ["09:00", "12:00"], "service_min": 5 /* ... 99 more stops */ ] # Response in ~8ms: "routes": [ "vehicle": "van_1", "stops": ["order_291", "..."], "eta_each": ["09:11", "..."], "total_km": 91.4 ], "unassigned": [], "compute_ms": 8, "cost_usd": 0.0003, "iterations": 6821
Rust-native ALNS with simulated annealing. Warm-start support for recurring daily routes.
Savelsbergh forward-slack timing arrays. O(1) feasibility checks. Hard and soft windows both supported.
Mark stops as in-progress mid-day and re-optimize the rest without touching locked assignments.
Every response returns iteration count, operator scores, and solution cost breakdown. No black box.
Click to add delivery stops. Drag to move them. Routes update instantly.
Heterogeneous fleets with per-vehicle capacity, shift hours, and speed profiles. Hard time windows enforced.
CVRPTWCustomer delivery windows with service duration per stop. Earliest departure, latest arrival, all exact.
VRPTWNew order dropped? Cancellation received? Re-solve in milliseconds around committed stops mid-day.
DYNAMICTested against CVRPLIB and Solomon VRPTW standard instances. Published gap-to-optimal. No marketing claims.
OPEN RESULTSRequest HTML/SVG output with any solve. Embed stop-by-stop maps in your ops dashboard or driver app.
SVG + HTMLPass a custom distance matrix (OSRM, Google, HERE) or let Raiden compute Euclidean. Zero map-provider lock-in.
FLEXIBLEBased on 1 CU minimum per solve (1-second time budget). Actual usage scales with time limit — longer solves produce higher quality routes.
No sales call. No per-stop shock. Start on the free tier and scale on compute alone.